Information Quality Is What Wins In The AI Era

Full analysis

Your draft

Tim Frank, Microsoft's corporate VP of Monetization, Commerce, and AI Economy, sat down with AdExchanger and made one argument worth taking seriously: the web now runs in three layers at once. A human web, an LLM-assisted web where people ask a model and then click, and a fully agentic web where the model buys for you. His pitch to brands is to restructure product data now so agents can find, recommend, and choose them. His slogan: "information is the new bid."

Here is what is actually being decided. Not whether agentic commerce is coming. Whether an operator should reallocate real budget and engineering time away from click-driven measurement and toward feed quality, structured product data, and agent-readable standards like the Universal Commerce Protocol. That is a mostly easy call to undo. Structuring your product data cleanly costs you nothing if agents never show up, and pays off if they do. The deadline, if there is one, is set by adoption curves nobody can see yet, not by a hard date.

Little of this touches programmatic plumbing directly. The MCP segment does. Let's separate the two.

The Market Analyst. Follow who benefits from this framing. Microsoft owns Bing, Edge, and Copilot, and it is pushing UCP into all three plus a Target Roundel loyalty tie-in. When a platform tells you the future runs on a standard it controls the surfaces for, that is a distribution play with a thesis wrapped around it. In plain terms: Microsoft wants to be the toll road for agent-driven shopping, so of course it says the toll road is the future. The read for public ad-tech: if discovery moves to agents, the companies exposed are the ones whose revenue depends on clicks and last-touch attribution. That is Criteo's retargeting core, that is a chunk of the open-web display stack. Retail media networks with clean first-party product feeds, Walmart Connect, Amazon Ads, Roundel, are better positioned because they already own the structured data agents want.

The Skeptic. Two of Frank's headline numbers do not survive a first read. Bot traffic "surpassing" human traffic at 60% is an old fact that was already old news before Frank cited it. Automated requests have been a huge share of raw web traffic for a decade, and most of it is scrapers and crawlers, not shopping agents deciding what to buy. The "8,000% growth in agentic browsing" figure comes with no base, no methodology, no publication. Eight thousand percent from near-zero is still near-zero. He credited HUMAN Security, a firm whose business is measuring bot traffic and therefore has every reason to make it sound enormous. Strip the numbers out and the thesis is still interesting. But the numbers are doing the selling, and they are not audited.

The Operator. Try to execute "information is the new bid" on Tuesday. You do not control your third-party reviews, your forum sentiment, or your return rates, which are exactly the signals Frank says agents triangulate on. You control your merchant feed. So the actionable part shrinks to: keep your feed accurate and complete, and stop blocking AI crawlers by reflex. That is genuinely good hygiene and mostly free. The harder problem shows up at 90 days. You cannot measure whether any of it worked. Frank admits click-through and traffic are dying as metrics and offers "citation frequency" and UCP conversion signals as replacements. Those live inside Microsoft's surfaces. You are being asked to optimize for a scoreboard the platform keeps.

The Customer / End User. For the buy-side operator, the MCP piece is the part that matters this week, and Mark, VP of Product at Basis, was honest about where it stands. MCP lets an agent talk to a DSP or SSP in plain language and pull reporting, build audiences, move budget, without a custom integration for each platform. Amazon, Meta, and Google have shipped MCP servers already. But Mark said the quiet part: letting one agent shift budget across platforms on its own is where people say the industry is going, not what anyone runs today. Real deployments are read-plus-single-platform-action. The gap between the demo and autonomous cross-platform buying is a governance problem, and most buy-side shops have written zero rules for what an agent is allowed to spend without a human.

The CFO. The cost of good product data is small and the payback is real either way, so that spend is easy to justify. The expensive mistake is believing the measurement shift is here and gutting your attribution stack to chase citation metrics. Frank's own timeline is "two years" and "should dwarf," which is a forecast, not a P&L. The disciplined move is to fund feed quality and MCP governance now because both are cheap and both are useful even if the agentic web stalls, and to keep the click-based measurement running until something audited replaces it. Do not pay a GEO/AEO vendor to game model outputs. Frank, who has every reason to want that market to exist inside legitimate optimization, flatly said the gaming version is a transient game the model teams are already closing.

Where the tension lives

The real disagreement is about timing and about who keeps score. Frank says the measurement stack is already breaking and offers Microsoft-native replacements. The Skeptic and the CFO say the evidence is a vendor stat with no methodology and a two-year forecast, so ripping out audited click measurement now is premature. Both can be right: the direction is real, the timeline is unproven, and the replacement metrics conveniently belong to the platform making the pitch.

The second tension is control. The Operator wants to act, but the signals that decide agent recommendations, reviews, return rates, third-party ratings, sit outside the brand's hands. "Information is the new bid" sounds like a lever you can pull. Most of it is a reputation you earned over years.

What it hinges on

Whether an operator should move now depends on one belief: does agent-driven discovery reach enough purchase volume to change where marketing dollars produce outcomes, and how soon. The council leans toward acting on the cheap parts and waiting on the expensive ones. Fix the feed, stop blocking crawlers, write agent-spend governance for your MCP integrations, all low cost and useful regardless. Do not tear down click-based measurement, do not buy GEO gaming services, and do not treat Microsoft's citation metrics as currency until something neutral can audit them.

Prediction: By the end of the 2027 upfront season (September 2027), no independent, audited cross-platform standard for measuring AI-agent-driven purchases will be in market, and the leading "agentic commerce" measurement signals will remain platform-native ones controlled by Microsoft, Amazon, or Google.

Confidence: Medium. The direction is real, but neutral measurement always lags the channel by years.

Why: Every new ad channel arrives with the platform keeping its own scoreboard first, and neutral audited measurement shows up only after buyers have enough spend at stake to demand it, which took years for CTV and retail media and is still incomplete there. Frank is already promoting citation frequency and UCP conversion signals, both of which live inside Microsoft's surfaces, and Amazon, Meta, and Google are shipping their own MCP servers with their own reporting. The parties who would build a neutral standard, the measurement vendors, are the ones the walled gardens least want holding the ruler, so the incentive runs against an independent standard emerging fast. The opposite outcome, a neutral audited agent-purchase currency by fall 2027, would require the platforms to invite outside grading of a channel they currently own end to end, which is not how any of them behaved with their existing walled inventory.

Revisit by 2027-09-30: We're right if agent-purchase measurement in market is still dominated by platform-native signals (Microsoft citation metrics, UCP conversion data, Amazon/Google MCP reporting) with no MRC-accredited or comparably independent cross-platform standard. We're wrong if an independent, audited standard for measuring AI-agent-driven purchases is live and adopted by at least one major holding company by then.